EEVblog Electronics Community Forum
Products => Test Equipment => Topic started by: Melt-O-Tronic on April 10, 2014, 05:38:51 pm
-
Is anyone else using correction factors? I've spent an obscene number of hours setting up a table to factor out ripple in a passive probe and keep running into problems saving the tables when there are more than about 50 points.
The UltraSpectrum software allows you to load up to 200 corrections, either hand-keyed or loaded from a CSV file. Then you can transfer those to the SA and they work fine. But if you save the table on the SA, the file cannot be recalled. Attempting to do so returns a dialog box that says, "Fail to load file".
It doesn't matter whether I save to internal memory or USB drive. It also doesn't matter whether the table was built manually through the DSA815's keypad (ugh!) or whether they're transferred from UltraSpectrum. I'm also not sure of the exact threshold where it stops working, but it's somewhere between 50 and 80 values. A table with 50 values can be saved and recalled with no problem, but 80 or more values can't. Also, "large" correction files saved to internal memory corrupt the directory display (screen shot attached).
I sent a PDF file and sample data files last week to Rigol NA with screen shots and concise instructions for recreating the issue, but haven't heard anything back yet. It seems a little early to poke them again. Am I the only one who has discovered this?
By the way, I first noticed this on firmware 00.01.07 and the condition persists in 00.01.08 (although in .08, none of my previously saved .07 files can be recalled!).
-
I have not used them, but would be interested to see what you are doing with them. Could you post some pics or a video showing how you determine the values and what the outcome of applying them is (if it is not a lot of work to do so...)?
-
I have an Auburn 3 GHz passive probe that I'm trying to apply corrections to. It has quite a bit of frequency-dependent response swing -- as much as 7.1 dB. A trace is attached below. Since the probe's effect on readings is highly repeatable, my goal is to have a correction file to zero out the ripple and attenuation of the probe, thus relieving my tiny little dinosaur brain from doing math. :)
Auburn says to just add 14.8 dB for 50-ohm circuits, but depending on frequency, I'm seeing as much as +2.8 dB to -4.3 dB error with that approach.
My method for characterizing the probe's response is to put an N-to-BNC adapter on the input and a short piece of coax (LMR-240UF) with an N connector on one end and BNC on the other on to the TG output. The cable then goes into the BNC adapter. I turn the TG on and normalize, disconnect the BNC, attach the probe to the input and directly probe the BNC on the end of the coax. I'm using the probe's narrowest ground pin which happens to match up to the BNC's dimensions pretty well.
So I then save the trace to CSV and get about 600 data points. I load that into Excel and calculate each point's sum of its absolute differences from its two immediate neighbors and sort descending by that. That way, I prioritize the readings according to how different they are from their neighbors. I then lop off the bottom 400 points, re-sort the remaining 200 by frequency and voila! I have the 200 most important correction factors.
With only 100 corrections, I can get the trace fairly flat, but any file I save with more than about 50 is lost forever because it can't be recalled. |O
-- Melt-O-Tronic / KB5EO / Destroyer Of Innocent Circuits :D
-
Thanks for the pic and explanation. I measured an Agilent 300 MHz 10X passive probe from my DSOX 2022A scope and got a really crummy looking response... I have not captured the data points to create a correction file yet, but I will give that a try when I get a chance to see if I get the same issue that you have.
Your Auburn probe is awsome by comparison! Is your probe a 1X? I see no noise on your trace and a lot on mine...
-
I think the difference in apparent noise is the vertical scale. Here's my Auburn over 300 MHz with 2dB/div like yours. The Auburn P-20A is a 10X probe.
I'd say you're in great shape if you want to build a correction table. You could flatten that whole curve with only a dozen entries. I'd suggest doing a 25x video average, freeze the trace and note the readings every 30 MHz plus the one at 15 MHz. Enter those values into the table (flipping the sign from negative to positive) and save the table.
Unfortunately, the number of ripples in the Auburn's response from 0 - 1.5 GHz means that I need a lot more corrections. I could save 4 different files for different frequency ranges, but that's clunky and it's frustrating that the Rigol has a bug that won't allow it to read in certain files that it created.
-
That's a nice passive probe you got there. :) Especially considering that there are bound to be some sneaky variations due to the TG in there, even after normalization.
Question: do you know what the limitations are regarding what kind of correction factor files the DSA815 will accept, and which ones it rejects for reasons known only to an overworked engineer at Rigol?
If you know that it should be doable to automate the procedure. Provided you don't mind installing python on a no-doubt-windoze machine. The idea would be to press button, get response from dsa, optimize curve to satisfy constraints for rigol file & minimize error under curve, save optimized compensation curve to file, use file on rigol. None of that excel faffing about.
You could even optimize it for use on different frequency spans. Anyways, since you clearly have played around with correction factors more than I have ... what are the specific constraints you have found out so far? (Other than "doesn't work, grrr" ;) )
-
mrflibble, you and I are on similar thought paths. I'm tinkering with VB to read the traces directly from the SA and generate correction tables. That would be nifty for initially creating the tables, but it's a big bother to need the PC to reload the values into the SA every time I want to use them -- especially if I take it out in the field as I'm likely to do.
The only limitations I know of are; 1) no more than 200 correction factors and 2) The DSA815 cannot recall any correction file if that file has more than ~50 values (the actual upper limit is somewhere between 50 and 80 values).
#1 is perfectly acceptable to me. #2 is not.
As far as sneaky variations due to the TG, that seems like a very real possibility, but I'm not sure how I could detect & compensate for that or whether they'd significantly affect the corrections. Of course, one obvious way to deal with it is to normalize through the probe, but the correction table to me seems like a much more convenient means of compensation. I'd like to have saved corrections for probes, directional couplers, etc.
One refinement I want to make to the algorithm for selecting the most meaningful corrections is this: For each value, compute the slope/intercept of a line directly between its neighbors, then calculate the point's distance tangent to that line and sort descending by that distance. The top N values are the ones that are retained for use in the table.
Wouldn't it be cool if we could write software that could run directly in the SA? 8) It's not gonna happen in this class of machine though.
-
software that could run directly in the SA?
RPi in a box velcroed to the back. Hit a button to upload whatever is stored on micro-SD, plug into a nearby HDMI screen to edit, etc.
-
I think the difference in apparent noise is the vertical scale. Here's my Auburn over 300 MHz with 2dB/div like yours. The Auburn P-20A is a 10X probe.
I'd say you're in great shape if you want to build a correction table. You could flatten that whole curve with only a dozen entries. I'd suggest doing a 25x video average, freeze the trace and note the readings every 30 MHz plus the one at 15 MHz. Enter those values into the table (flipping the sign from negative to positive) and save the table.
Unfortunately, the number of ripples in the Auburn's response from 0 - 1.5 GHz means that I need a lot more corrections.
Are you sure those ripples are not caused by poor grounding, impedance mismatch or something like that? Auburn's website shows a much flatter response for that probe.
-
My response curve is profoundly repeatable, day after day, even with everything else turned off. I interpret the graph on the Auburn site to mean, "this is roughly what the response looks like."
(http://www.auburntec.com/auburnchart2.jpg)
Regardless, the Rigol DSA815 has a bug and that's the real point of the thread. ;)